In modern Silicon Valley, free food for employees is pretty much table stakes, but when Google started offering it more than two decades ago, it was a new and opulent perk. Now, the tech giant is doing something novel with the stuff employees don’t eat: it’s training AI.
On Thursday, Google announced a partnership with Mill — maker of a $1000 “smart” trash bin — to use a dataset Google created from its own food waste, which it labeled and annotated years ago for computer vision research. Under the terms of the deal, Mill will also get early access to unreleased versions of Google’s flagship Gemini AI models, as well as a team of its AI engineers and researchers.
Mill is the second startup from Matt Rogers, a Nest cofounder who joined Google after the company bought Nest in 2014. He left and co-founded Mill in 2020 with a single obsession: food waste. Mill’s high-tech trash can processes food scraps into chicken feed, and, for a fee, ships it to farms. The startup sells the trash cans to both consumers and businesses, like grocery stores, restaurants and offices, which can deploy the bins in their kitchens and cafeterias across locations. Amazon-owned Whole Foods, for example, was announced as a corporate customer in December.
The idea is that by knowing what food is in your trash — via a camera inside the bin — businesses could make better decisions about their procurement. If a company’s catered lunches reliably end with pounds of macaroni and cheese in the trash, the company can order less. Or if a grocery store sees that much of the Caesar salad from its prepared food section is routinely discarded on Tuesday and Thursday nights, it could adjust its menu accordingly. In theory, tracking like this can also make it easier to route surplus to food banks and document donations for tax breaks. “No one is going to sort and sift through this stuff and go to the CFO to say, ‘Let’s do something different,'” said David Krane, CEO and managing partner of Alphabet’s venture arm GV, and one of Mill’s earliest backers.
The food waste dataset comes from Project Delta, an effort that started inside Google’s X moonshot division. In 2017, the company decided to analyze its food scraps by creating a camera and software system in its kitchens. Researchers then identified and labeled the scraps, as well as outlined what was food and what wasn’t. Then they showed the data to chefs to review the results. The goal was to help grocers, logistics companies and food banks match food supply with food demand, like an “air traffic controller for food,” Google wrote at the time.
Project Delta never became a standalone business. Google instead folded it into the broader company in 2020 and repurposed some of its research for a supply chain effort called Project Chorus. But the food waste data was just being …wasted. “We’re kind of the inheritors of that project,” Rogers told Forbes.
The goal is to cut down on food waste altogether, which totals about 30% to 40% of the food supply in the U.S. each year, according to the USDA. That number adds up to 70 million tons, and tallies to $380 billion in economic value, per a report published earlier this month by ReFed, a Chicago-based nonprofit focused on studying food waste. Rogers argues this is a problem AI can fix — and one that should seem absurd in hindsight. “If you look into the future, like 10 or 20 years, I think we would be embarrassed if we were still wasting as much food as we are,” he said. “This is something that technology and design can help solve.”
The Mill partnership is through a Google program called AI Futures Fund, where the tech giant partners with startups to provide access to its AI tools, in exchange for feedback on the models’ performance. Other partners include legal tech startup Harvey and enterprise search engine Glean. Google says Mill is the first hardware company to join the program.
During the interview, conducted over a video call, a garbage truck passed by outside, loudly collecting waste from bins. As if right on cue, Rogers noted the commotion. “Wouldn’t it be great if a garbage truck didn’t come every day?” he said. “That’s kind of the idea. It’s gonna take us a while to get there. But could they come once a month? I think we could get there.”
